Give an agent live web results without a search contract

Four requests on one path prefix, each its own priced call: search runs a neural or keyword query over the live web with an optional category filter and domain include/exclude lists; find-similar returns pages semantically near a URL you supply; contents extracts clean text from a list of URLs and is priced per URL rather than per call; answer returns a grounded answer with the sources it drew on.

A typical agent loop is search, then contents on the handful of results worth reading, then answer — three requests, three prices, each quoted before it runs. The parameter table below is the search request the endpoint validates.

In Claude Code, Claude Desktop, Cursor, Windsurf, Cline and any other MCP host, this is the blockrun_exa tool of the BlockRun MCP server — the agent calls it like any other tool and the result comes back in the conversation. In OpenClaw, Codex and Hermes, ClawRouter exposes the same catalog. In your own code, the Python and TypeScript SDKs wrap the endpoint above with types; any HTTP client works too. Every harness, with its setup →

No account and no API key stand between an agent and the first call: the request is quoted in dollars and the agent's client pays it. A team that prefers keys gets one, with a monthly invoice.

An agent that needs the live web usually picks one of four things: a SERP scraper (SerpAPI and its peers), which returns what Google shows; a search API built for LLMs (Tavily, Brave Search API), which returns pages and snippets shaped for a prompt; a neural index (Exa), which returns pages by meaning; or a full search-and-read pipeline the agent assembles itself. Each needs its own key and its own monthly plan.

This endpoint is the neural index — Exa's — with keyword search, find-similar, content extraction and a grounded answer on the same path prefix, each priced per request with no plan. A typical loop is search, then contents on the handful of results worth reading, then answer: three requests, three prices, each quoted before it runs.

What it is not: a Google results scraper. If the job is "what does the first page of Google say", a SERP API is the right tool. If the job is "find pages about this, read them, and answer", this is.